Building a Venture Scale Enterprise AI Company from Spain: David Villalon, CEO of Maisa (Part 1)

David discusses his journey building an enterprise AI venture with 15 large enterprise customers who are implementing real use cases built on the Maisa platform. These use cases are not POC experiments. They are in production. The customer base currently is mostly European enterprises. This is going to be a very valuable company.

Sramana Mitra: All right, David, let’s start at the very beginning of your journey. Where are you from? Where were you born, raised? What kind of background?

David Villalon: I’m from Spain as you can probably notice by my accent. I come from a very technical background. I have two bachelors — computer engineering and business administration — and this is my second company. I am living in Valencia, and I’ve always lived here.

I dabble a bit in magic. I’ve always been very involved in different ventures. I always like to try new things. I was very lucky to join a company that was a real-time AI voice changer back in 2019. That was already on the first wave of AI.

And I was Director of Product, which taught me a lot about how to scale products to 72 million users. And it was crazy.

Sramana Mitra: What company did you work for?

David Villalon: Voicemod. It’s a real-time AI voice changer, especially for the content creator space and streaming or gaming, Roblox, Discord.

We scaled during 2020 and 2021, and it was a crazy experience to see what it takes to grow that fast. I was very lucky.

Sramana Mitra: Remotely?

David Villalon: Well, no. I worked in person and remotely. Remotely because we had COVID; in person because it was in Valencia. When I joined, we were 14 people, and in one year and a half we grew to about 200. It had tremendous success. We raised a Series B for $40 million. I was the Director of Product at that moment, and it was really, really cool.

The good thing is that I was very lucky because I suddenly saw one model that was called GPT-3 and a company that wasn’t really well known at that time — OpenAI. I decided to test a few of the models that they had and started to play around with AI. It was like, “Wow, what is this?” Because it was a playground. It was very different.

So, I decided to build a chatbot in January or February of 2022. It was released in November, I think.It blew my mind because I started to see the potential. So it was very early. I started experimenting, building functionalities while working there, and got hooked by AI.

Sramana Mitra: Were you still working in the previous company?

David Villalon: No — um, no, no. I was still working, yes. At that moment I was still working at that company, but I was doing that in my free time.

Sramana Mitra: Okay.

David Villalon: So then, in the next year, I got the opportunity of being Chief AI Officer of a company. They’re going to pay me for what I do in my free time, which is playing with models and with AI. That was in 2023.

So, I decided to take the opportunity. The Chief Scientist Officer of that company is now my co-founder. We were working on foundational AI. We were both working on making models for the Spanish market, especially at the foundational level of AI. That’s when we started to see all the opportunities.

Sramana Mitra: What kind of company was it that you became the Chief AI Officer of?

David Villalon: It was called Clidrive, with a transition to Clibrain. The first one was a company that was for renting cars. So we started to use AI to make the call center work at 10x, to make the internal sales team work at 10x. That’s when we decided that there was an opportunity. This was early 2023 — and to found a company just for the Spanish market, because there was a huge gap there at that moment.

My co-founder and I then decided to found our own company. It was really good because we gained a very contrarian view of the market. That led to Maisa appearing because we had firsthand insights about what the problems really were.

We thought the solutions the market was proposing weren’t the right ones.

Sramana Mitra: So, what was the problem that you decided to go after in Maisa?

David Villalon: The problem was that it was very hard to make AI work for a specific use case. Now it’s easier, but it’s still hard. It took a lot of effort and time, because AI has intrinsic limitations — things that you cannot remove from it. It hallucinates, it’s not reliable, it’s absolutely unfaithful. You cannot trust what it’s telling you.

No matter how much you try, you really cannot tell why things are not working. So, you enter into never-ending loops of trying and testing and iterating. It was very frustrating for us.

This lack of transparency and the inability to trust the results was supposedly solved by the market with Retrieval-Augmented Generation (RAG). We believed at that moment that this was also wrong, by positioning ourselves into the future.

It’s very hard for us to build with AI. So, we did not start with a problem from a user directly — we were the users who hated building with AI, even though we both came from that market.

We saw that in the future, you’re going to have models that go very fast, and that AI does in an hour what would take weeks. And how are you going to know if it’s well done if you cannot trust what it has done?

So, we decided to use AI very differently and investigate: What if we don’t use AI to get responses, make a summary, or generate text? What if we use it like an operating system? What if we use it to tell us what needs to be done by a computer — like executed, printed, written — to get to the response?

Then, I don’t really need to trust what the AI is telling me, because I have the proof of work, which is what has been executed to get the result.

Trying to simplify it a lot — that’s what we decided to investigate.

We called that technology KPUA — Knowledge Processing Unit and Agent Reasoning Engine. And it worked — much better than expected.

This segment is part 1 in the series : Building a Venture Scale Enterprise AI Company from Spain: David Villalon, CEO of Maisa
1 2 3